I attended a session at the latest storage conference that gave some
specific performance suggestions regarding AIX/TSM/FastT. Maybe these will
help with your testing, I haven't had a chance to try all of these yet. I
have always used RAID5 on a FastT with very acceptable performance (due to
the write cache). Standard disclaimer applies, research these values and
implement at your own risk.
FC Adapter Attribute Changes:
lg_term_dma from 0x200000 to 0x1000000 (from 2MB to 16MB)
num_cmd_elems from 200 to 1024 (queue depth for the adapter)
dar attribute changes:
load_balancing was set to yes
autorecovery was set to yes
hdisk attribute changes:
queue_depth from the default to 16 (total for all hdisks on a FastT should
be 512 or less)
prefetch_mult from 0 to 8
AIX vmtune changes:
minperm from 20 to 5
maxpgahead from 8 to 32
maxrandwrt from 0 to 32
numclust from 1 to 8
sync_release_iolock from 0 to 1
maxclient from 80 to 5
General FastT disk recommendations:
RAID5 is ideal for the recovery log
Create LUNs with 2 PVs, one on each half (RAID 1 pair)
Storage Pool drives have segment size of 128K, read/write cache enabled,
write cache w/mirroring, cache read ahead 4
Database drives have write cache enabled, write cache w/mirroring, cache
read ahead 0
______________________________
John Monahan
Senior Consultant Enterprise Solutions
Computech Resources, Inc.
Office: 952-833-0930 ext 109
Cell: 952-221-6938
http://www.compures.com
Leonard Lauria
<leonard AT UKY DOT EDU>
Sent by: "ADSM: To
Dist Stor ADSM-L AT VM.MARIST DOT EDU
Manager" cc
<[email protected]
.EDU> Subject
opinions on disk partitioning
08/21/2003 07:53
AM
Please respond to
"ADSM: Dist Stor
Manager"
<[email protected]
.EDU>
I would like to get opinions on disk configurations for a new platform.
I am installing my TSM server on an aix platform, with FASTT700 disk.
The FASTT700 has (28) 73 GB, 15K disks, of which 23 are available for this
application.
I am considering the following 2 setups, but please feel free to make other
suggestions!
1) 12 drives for the TSM DB, raid 10. I only need 100 GB, plus room to
grow,
perhaps double, so this wastes a lot of space, but I get needed
spindles.
Remaining 11 drives would be Raid 5, and exported as a single LUN
which would
have 4 logical volumes for:
570 GB disk pool
200 GB disk pool
20 GB disk pool
13 GB TSM log
2) 8 drives for the TSM DB, raid 5. I only need 100 GB, plus room to
grow,
perhaps double, so this is less space, but fewer spindles and no
mirror.
Remaining 15 drives would be Raid 5, and exported as a single LUN
which would
have 4 logical volumes for:
500 GB disk pool
500 GB disk pool
82 GB disk pool
13 GB TSM log
#2 gives me more spindles for the disk pools, by using only raid5 for the
database partitions.
Another concern is having the disk pools compete with each other on the
same disks. Would
it be better to have fewer spindles per disk pools, but have disk pools
seperate from each other,
or all the disk pools spread over the same larger number of spindles?
Also, I have had a lot of conflicting information regarding TSM doing
mirrors of the DB and LOG
versus letting the FASTT hardware do raid protection. It seems the
hardware implementation would
be faster, and just as safe, as letting TSM do mirrors...not to mention
allowing me to spread things
out a bit more.
Any opinions on my options?
Thanks!
leonard
|
